Recap WITI Orange County - February 17

Terry Dear

  • Share:    
Can Machines Think? Introduction to Machine Learning and Artificial Intelligence
February 17, 2016
Recap by Terry Dear

In the book, 2001: A Space Odyssey, written in 1968, HAL (Heuristically programmed Algorithmic computer) controls the books' spacecraft and interacts with the ship's astronaut crew. HAL is capable of speech, speech recognition, facial recognition, natural language processing, lip reading, art appreciation, interpreting emotional behaviors, automated reasoning, and playing chess.

Fast forward, 2016....the future is NOW. HAL has become a woman....and our WITI Orange County attendees got a chance to meet Amelia (from IPsoft), Geminold F. (first robot actress), and experience Fresh Dress, a concierge like experience that combs through thousands of dress images on the Internet to find just the perfect mid-thigh, no sleeves, red dress for business and evening wear.

How is that done? The rest of our evening was spent discussing machine learning use cases, terminology, and trends with Jonathan Crane, Chief Commercial Officer at IPsoft (http://www.ipsoft.com/), Jennifer Bolton, Sr. Director of Marketing at Dato (https://dato.com/), and Dr. Srikrishna Sridhar, Data Scientist at Dato.

One of the most interesting juxtaposition is the polar shift in relationship between technology and labor, representing the automation of the knowledge worker.

Old paradigm: 'Technology supports Labor' - where computers assist the humans who do the work and make decisions.

New paradigm: 'Labor supports Technology' - that is, humans are now assisting the computers who are making decisions and doing the work.

Artificial intelligence was what it was called in HAL days. Today, machine learning is considered beneficial intelligence. When a smartphone can 'see' and guide a blind person through traffic, that level of intelligence is not just artificial, but beneficial, like self-driving cars.

How is it that beneficial intelligence is here now? Several factors paved the way with the 21st century of electrification, affordable computers, speed of change beyond the human capacity, mobility, internet, IOT, cloud, social media, and virtual versus physical hardware/software.

Let's get back to Amelia. Jonathan shared with us that Amelia was 'birthed' from Eliza. (Eliza being the Computer Therapist learning about your feelings - the first artificial intelligent program from the 60's.) After thousands of studies, HAL also became a woman. It was discovered that a woman is better received by males and females in empathy quotient and interaction. Amelia is designed to be your digital employee. She has memory, contextual comprehension, and emotional responsiveness. She can understand what is meant and not just the words. She continues to learn and solves problems. She is programmed where her natural language is mapped to business processes. She has 'episodic memory' where she gathers all inputs to handle a process. She understand context switching. She can integrate with SAP (ERP backend systems). Most of all, she has emotional intelligence to match facial expressions to empathy as to what is being said. Visit the IPsoft website to understand more of the technology driving Amelia. One takeaway is that Amelia is fallible. Amelia learns from the human agents. And humans can make mistakes. In this new world, there will need to be a new role something like 'digital compliance manager' who evaluates what Amelia will really learn. And yes, Amelia falls short of HAL as she does not lip read or been programmed to appreciate art.

De-mystifying machine learning more, Jennifer and Srikrishna talked about the chicken and the egg. In order for machines to learn (and be accurate), how much data do they need? A small data set will yield different decisions than a large data set. It was stated that the minimum viable model size is around 10,000 data points. A deep learning model works with one million data points. Jennifer and Srikrishna walked us through the dataset, use cases, and toolkits.

The first component is data: what kind of data do you have? Images? Text? Transactions? Or Customer Usage? After data, what types of business use cases are machines best at doing (replacing the human)? The most frequent uses (today) are for:

(1) Making recommendations: Think Pandora or Netflix, the ability to increase user engagement to stay on your site longer with personalization.
(2) Identifying customer churn: identifying patterns so that you do not lose your customers but can keep them, as well as optimize your marketing spend.
(3) Noting fraud detection: Think credit card providers, how are they able to note a fraudulent credit transaction by noting spending behavior.
(4) Doing image analysis: the speed of sorting and tagging similar images is amazing, finding that needle in a haystack. Are you looking for the perfect red dress on the internet?
(5) Lead scoring: predict and customizes a prospects experienced based on factors such as interest level, experience, etc.
(6) Performing sentiment analysis: Eliminating the need for user to read everything, the machine will review blogs and aggregate the tone/flavor/sentiment of the postings providing summarizations. Best example are product reviews.

Based on the above use cases, can you guess which industries are making headway adopting machine learning? Not surprisingly they are: e-commerce sites, followed by SaaS, online media, government, financial services, and then health care.

How practical is it for me to apply machine learning in my organization? Dato has taken the machine learning from the academic world of algorithms (describing all the business rules/neural networks/etc.) to toolkits, making machine learning a reality for organizations. Dato developed these toolkits against large datasets to provide startup companies the same quality of machine learning as large corporations. To learn more, visit the Dato website and get started. If you know Python, you can get started today.

And if you want to be part of the beneficial intelligence revolution, attend the Data Science Summit in San Francisco, CA. Enter the special discount code "WITI20" for a 20% discount off the Data Science Summit ticket price through June 30, 2016. The 2 days of hands-on training and talks is July 12-13, 2016 in San Francisco. Click http://conf.dato.com/2016/us/ for more information.

A special shout out to Solugenix (http://www.solugenix.com) for sponsoring our event. Solugenix is an IT consulting, professional staffing, managed services organization where over 50% of their employees are women.

Good Night, Amelia.

Tags

Comments